HDD weirdness using HDPARM

[Monoman]

I normally try to help as often as possible, but I to do have a question now and again


well, this is a little weird to me, I have tested this on 3 computers and I have gotten 2 different results on the 3(2 matching and 1 is different)


here is the odd ball score:

hdparm -tT /dev/sda
beffer-cache reads 128mb 0.61sec = 209.84
beffered disk reads 64mb 1.86sec = 34.41

Here is the score thats identical:

hdparm -tT /dev/sda
beffer-cache reads 128mb 0.32sec = 490.91
beffered disk reads 64mb 1.86sec = 34.41

I am sure you noticed the difference. What I have is an adaptec 29160 with a U160 10k IBM drive. one computer I used this on is a p4 1.4 and a pIII 800. The odd ball score is on an AMD 1.4 (which is overclocked and when I clock it down it makes no difference) How I test is I run the Slackware 9.1 install cd and when I got to the command prompt, I run the bench. now I have had a weird issue with my MB that no one can answer and that is the fact that all of my peripherals are all on the same IRQ(there are about 10 of them from usb, firewire, sound, modem, nic, ect..) is it possible there is an issue with the MB and the SCSI card? or could it just be my worrying about this.

The main reason I ask is because I am upgrading to a 15k.3 Cheetah and for it to run as expected, I don't want that AMD rig to be a problem.(which is it's intended home)

Thanks for reading!

Mitch

 

[Monoman]

up anyone?

 

[Monoman]

up

 

[Monoman]

up

 

[TheBoyBlunder]

Do you have an ACPI compliant AMD box?

 

[drag]

The buffer cache reads are not accurate anyways.

The differences can be accounted for different chipset designs and drivers, but even then the buffer cache reads do not reflect any real-world performance difference. It's more of a way to do relative comparisions on the same computer.

Use something like iozone to compare relative performance. HDPARM is a tweaking tool, more then a way to compare performance. It can, but only in a relative way, not a definate way.


Also check out and compare the differences between:
dmesg output,
hdparm /dev/hda
hdparm -i /dev/hda

 

[Monoman]

Thanks drag, I was aware they are not a very good measurement of performance, but I was just asking because of the difference between computers as 2 got identical scores and the third scored lower..

 

[InlineFive]

Originally posted by: Monoman
Thanks drag, I was aware they are not a very good measurement of performance, but I was just asking because of the difference between computers as 2 got identical scores and the third scored lower..

It's probably dependant somewhat on the chipset. Does the 3rd computer have a different southbridge then the other two?

-Por

 

[Monoman]

the two identical scoring systems are a pIII 800 and a P4 1.4 with an Intel Southbridge. The slower preforming system in an AMD 1.4 with the VIA southbridge.

 

[Monoman]

come to think of it, every ide or scsi drive gets the same buffer-cache-read score... is the SB the bottle neck?

 

[Netopia]

It's not just the SB, it's the memory subsystem in general. I have a box with an Abit KT7-A Raid (KT133a chipset) and I can get results for buffered cache reads (why did you write bEffered?) anywhere from about 120-240 MB/s depending on how I have my MEMORY settings in the BIOS. Try it out and you'll see what I mean.

Joe

 

[Monoman]

LOL your the first on two forums to notice my typo.... nice work. I have the KT133a chipset also.. wierd... that a bottle neck, compaired to my older PIII system.

Thanks

 

[Netopia]

LOL your the first on two forums to notice my typo.... nice work.

That's funny... when I was responding, I wondered "is he doing that as a test to see if anyone catches it"?

Here's what I find weird....

Your KT133a scores are fine for a KT133 chipset computer.

Your P3 Scores are GREAT.

Your are VERY low compared to what I get.

For instance, here are the scores I get from a Dell I got off of hot deals. It's got a P4-2.8GHz and PC 3200 RAM


[root@xfer sbin]# ./hdparm -tT /dev/hda1

/dev/hda1:
Timing buffer-cache reads: 128 MB in 0.19 seconds =673.68 MB/sec
Timing buffered disk reads: 64 MB in 1.41 seconds = 45.39 MB/sec

That's off of a WD 80 Gig SE


This is my KT133a chipset. Athlon 1.2 with PC133 at moderate settings. 2 IBM GXP120's on a striped software raid partition and then from a single drive:

{Raid}
[root@cranium sbin]# ./hdparm -tT /dev/md0

/dev/md0:
Timing buffer-cache reads: 128 MB in 0.63 seconds =203.17 MB/sec
Timing buffered disk reads: 64 MB in 0.78 seconds = 82.05 MB/sec

{Single Drive}
[root@cranium sbin]# ./hdparm -tT /dev/hde1

/dev/hde1:
Timing buffer-cache reads: 128 MB in 0.62 seconds =206.45 MB/sec
Timing buffered disk reads: 64 MB in 1.40 seconds = 45.71 MB/sec



What I want to know is HOW DID YOU GET SUCH KILLER SCORES FROM YOUR P3? I know that the P4 had a quad-pumped bus, so that I understand, but how did you manage such nice bEffered scores on your P3?

Joe